RIVA 128
Nvidia's first integrated 2D/3D accelerator that rivaled Voodoo Graphics.
The RIVA 128, also known by its internal codename NV3, hit the market in 1997 as Nvidia’s first consumer graphics chip to combine 3D acceleration with built-in 2D and video support. Its full name stands for Real-time Interactive Video and Animation accelerator. This product came after Nvidia’s earlier and far less successful NV1 chip, and it marked the company’s first real breakthrough in public recognition, as well as a sharp pivot in its technical approach.
By 1996, Nvidia was in serious financial trouble. Its initial chip, the NV1, had relied on a unique rendering method called quadratic texture mapping. After that failure, the company spent a year trying to develop a graphics chip for Sega’s Dreamcast console, but Sega eventually canceled the project and switched to another vendor. However, Sega did invest $5 million to keep Nvidia afloat. Nvidia then laid off half of its 100 employees and poured all remaining resources into the RIVA 128. By the time the chip launched in August 1997, the company had only one month of payroll left. This crisis gave rise to what is still the unofficial company motto: “Our company is thirty days from going out of business.”
The RIVA 128 was built to work with Direct3D 5 and OpenGL, and it was specifically optimized for Direct3D—a major departure from the NV1, which had used quadratic texture mapping and didn’t support Direct3D at all. The chip contained 3.5 million transistors, was manufactured on SGS-Thomson’s 350 nm process with five metal layers, and ran at 100 MHz. It had a single pixel pipeline that could output one pixel per clock when sampling one texture, with a claimed fill rate of 100 million pixels per second and 1.5 million 25-pixel triangles per second. On-chip memory totaled 12 KiB, split between pixel and vertex caches. In 3D mode, the chip was limited to 16-bit color (highcolor) and a 16-bit Z-buffer.
The 2D engine was 128 bits wide and also ran at 100 MHz. Nvidia called this a “fast and wide” design, and it performed well for GUI acceleration compared to rivals. A 32-bit hardware VESA-compliant SVGA/VGA core was included, along with video acceleration optimized for MPEG-2 (though not full acceleration). Output went through an integrated 206 MHz RAMDAC. Unlike 3Dfx’s Voodoo Graphics, which required a separate 2D card, the RIVA 128 handled both 2D and 3D in a single chip.
Quick Facts
- Codename
- NV3
- Created
- 1997-08-25
- Highend
- RIVA 128, RIVA 128 ZX
- Openglversion
- 1.0
- Predecessor
- NV1
- Successor
- RIVA TNT
- Support Status
- Unsupported
- Manufacturer
- SGS-Thomson, TSMC
- Designfirm
- Nvidia
Facts from the source article.
Lore & Background
By 1996, Nvidia was in poor financial shape. It had initially pursued quadratic texture mapping with its NV1 product, then spent a year trying to build a graphics chip for Sega's Dreamcast. Sega pulled the plug on Nvidia's project but invested $5 million to keep the company alive. Nvidia laid off half of its 100 employees and focused remaining resources on developing the RIVA 128. By its release in August 1997, Nvidia was down to one month of payroll, leading to the unofficial company motto: 'Our company is thirty days from going out of business'.
The RIVA 128 was built to render within Direct3D 5 and OpenGL API specifications, a departure from the NV1's quadratic texture mapping. It consisted of 3.5 million transistors on a 350 nm process, clocked at 100 MHz, with a single pixel pipeline capable of 1 pixel per clock. It had 12 KiB of on-chip memory for pixel and vertex caches, and was limited to 16-bit color and 16-bit Z-buffer in 3D mode. The 2D engine was 128 bits wide, and the chip included a 206 MHz RAMDAC. It used 4 MiB of SGRAM on a 128-bit bus, providing 1.60 GB/s bandwidth in a unified memory architecture.
In early 1998, Nvidia released the RIVA 128 ZX, which increased memory support to 8 MiB and RAMDAC frequency to 250 MHz. The ZX version was fabricated by SGS-Thomson and TSMC, using an 8 MB SGRAM chip from Samsung Electronics clocked at 125 MHz. The original RIVA 128 had an AGP 1X interface, while the ZX was one of the early AGP 2X parts.
Reader's Guide
The RIVA 128 was notable as one of the first combination 2D/3D cards that could rival 3Dfx Voodoo Graphics. Unlike Voodoo, which required a separate 2D card, the RIVA 128 offered a single-card solution for both 2D and 3D, lowering system cost. Its 2D capability was competitive with high-end 2D-only cards. The unified memory architecture allowed 3D resolutions of 800×600 and 960×720, higher than Voodoo's 640×480.
Image quality was initially criticized. The card used per-polygon mipmapping instead of per-pixel, causing texture detail to 'pop' into place. Its bilinear filtering was sharper than Voodoo's but added light noise to textures due to a lower-fidelity algorithm, and there were noticeable seams between polygons. Later drivers addressed these issues, offering per-pixel mipmapping, full-scene anti-aliasing (supersampling), and image quality arguably matching Voodoo Graphics.
Driver support was rough for a significant portion of the card's life. Direct3D issues were apparent, and OpenGL support was initially poor. Many games used 3Dfx's proprietary Glide API, which only 3Dfx cards could legally use. The RIVA 128 could not accelerate the Unreal Engine in Direct3D mode due to missing hardware features, though OpenGL rendering was possible but slow and buggy. Nvidia's final Windows 9x drivers included a full OpenGL driver requiring 16-bit desktop color depth. Windows 2000 and XP included a driver lacking 3D support; a beta OpenGL driver was leaked but canceled, and no Direct3D support existed for those systems.
Did You Know?
- The RIVA 128 was Nvidia's first product to integrate 3D acceleration with 2D and video acceleration.
- Nvidia was down to one month of payroll when the RIVA 128 was released in August 1997.
- The RIVA 128 used a unified memory architecture, sharing the entire 4 MiB RAM pool between framebuffer and texture storage.
Frequently Asked Questions
What is the RIVA 128?
It was Nvidia's first consumer-facing GPU that merged 3D rendering, 2D acceleration, and video playback into a single chip. Released in 1997 under the internal codename NV3, it succeeded the poorly received NV1 and became the company's first widely recognized product.
What does the name 'RIVA' actually stand for?
The acronym expands to 'Real-time Interactive Video and Animation accelerator,' which neatly summarizes the chip's three-in-one design philosophy of handling gaming, desktop graphics, and video all at once.
How did the RIVA 128 stack up against 3dfx's Voodoo Graphics?
It was positioned as a direct rival to Voodoo Graphics, delivering comparable 3D performance while adding the 2D and video features that Voodoo simply lacked. That integrated approach meant buyers no longer needed a separate 2D card, a meaningful selling point in 1997.
What were the RIVA 128's key hardware specifications?
The chip was fabricated on SGS-Thomson's 350 nm 5LM process, contained roughly 3.5 million transistors, and ran at a 100 MHz core clock paired with 4 MiB of SGRAM clocked at the same 100 MHz speed.
Why does the RIVA 128 matter in Nvidia's history?
It rescued the company from the financial crisis that followed the NV1's commercial failure and established Nvidia as a legitimate GPU vendor virtually overnight. The chip also marked a fundamental pivot away from NV1's quadratic tessellation method toward a more conventional rendering pipeline that the broader industry could actually compete with.
More in Nvidia and AMD Graphics Processors 1-24
Spotted an error? Know more?
Reader corrections go straight into our review queue. Suggest an edit · How this site is sourced